mnist.py

from .vision import VisionDataset
import warnings
from PIL import Image
import os
import os.path
import numpy as np
import torch
import codecs
import string
import gzip
import lzma
from typing import Any, Callable, Dict, IO, List, Optional, Tuple, Union
from urllib.error import URLError
from .utils import download_url, download_and_extract_archive, extract_archive, \
    verify_str_arg


class MNIST(VisionDataset):
    """`MNIST <http://yann.lecun.com/exdb/mnist/>`_ Dataset.

    Args:
        root (string): Root directory of dataset where ``MNIST/processed/training.pt``
            and  ``MNIST/processed/test.pt`` exist.
        train (bool, optional): If True, creates dataset from ``training.pt``,
            otherwise from ``test.pt``.
        download (bool, optional): If true, downloads the dataset from the internet and
            puts it in root directory. If dataset is already downloaded, it is not
            downloaded again.
        transform (callable, optional): A function/transform that  takes in an PIL image
            and returns a transformed version. E.g, ``transforms.RandomCrop``
        target_transform (callable, optional): A function/transform that takes in the
            target and transforms it.
    """

    mirrors = [
        'http://yann.lecun.com/exdb/mnist/',
        'https://ossci-datasets.s3.amazonaws.com/mnist/',
    ]

    resources = [
        ("train-images-idx3-ubyte.gz", "f68b3c2dcbeaaa9fbdd348bbdeb94873"),
        ("train-labels-idx1-ubyte.gz", "d53e105ee54ea40749a09fcbcd1e9432"),
        ("t10k-images-idx3-ubyte.gz", "9fb629c4189551a2d022fa330f9573f3"),
        ("t10k-labels-idx1-ubyte.gz", "ec29112dd5afa0611ce80d1b7f02629c")
    ]

    training_file = 'training.pt'
    test_file = 'test.pt'
    classes = ['0 - zero', '1 - one', '2 - two', '3 - three', '4 - four',
               '5 - five', '6 - six', '7 - seven', '8 - eight', '9 - nine']

    @property
    def train_labels(self):
        warnings.warn("train_labels has been renamed targets")
        return self.targets

    @property
    def test_labels(self):
        warnings.warn("test_labels has been renamed targets")
        return self.targets

    @property
    def train_data(self):
        warnings.warn("train_data has been renamed data")
        return self.data

    @property
    def test_data(self):
        warnings.warn("test_data has been renamed data")
        return self.data

    def __init__(
            self,
            root: str,
            train: bool = True,
            transform: Optional[Callable] = None,
            target_transform: Optional[Callable] = None,
            download: bool = False,
    ) -> None:
        super(MNIST, self).__init__(root, transform=transform,
                                    target_transform=target_transform)
        self.train = train  # training set or test set

        if download:
            self.download()

        if not self._check_exists():
            raise RuntimeError('Dataset not found.' +
                               ' You can use download=True to download it')

        if self.train:
            data_file = self.training_file
        else:
            data_file = self.test_file
        self.data, self.targets = torch.load(os.path.join(self.processed_folder, data_file))

    def __getitem__(self, index: int) -> Tuple[Any, Any]:
        """
        Args:
            index (int): Index

        Returns:
            tuple: (image, target) where target is index of the target class.
        """
        img, target = self.data[index], int(self.targets[index])

        # doing this so that it is consistent with all other datasets
        # to return a PIL Image
        img = Image.fromarray(img.numpy(), mode='L')

        if self.transform is not None:
            img = self.transform(img)

        if self.target_transform is not None:
            target = self.target_transform(target)

        return img, target

    def __len__(self) -> int:
        return len(self.data)

    @property
    def raw_folder(self) -> str:
        return os.path.join(self.root, self.__class__.__name__, 'raw')

    @property
    def processed_folder(self) -> str:
        return os.path.join(self.root, self.__class__.__name__, 'processed')

    @property
    def class_to_idx(self) -> Dict[str, int]:
        return {_class: i for i, _class in enumerate(self.classes)}

    def _check_exists(self) -> bool:
        return (os.path.exists(os.path.join(self.processed_folder,
                                            self.training_file)) and
                os.path.exists(os.path.join(self.processed_folder,
                                            self.test_file)))

    def download(self) -> None:
        """Download the MNIST data if it doesn't exist in processed_folder already."""

        if self._check_exists():
            return

        os.makedirs(self.raw_folder, exist_ok=True)
        os.makedirs(self.processed_folder, exist_ok=True)

        # download files
        for filename, md5 in self.resources:
            for mirror in self.mirrors:
                url = "{}{}".format(mirror, filename)
                try:
                    print("Downloading {}".format(url))
                    download_and_extract_archive(
                        url, download_root=self.raw_folder,
                        filename=filename,
                        md5=md5
                    )
                except URLError as error:
                    print(
                        "Failed to download (trying next):\n{}".format(error)
                    )
                    continue
                finally:
                    print()
                break
            else:
                raise RuntimeError("Error downloading {}".format(filename))

        # process and save as torch files
        print('Processing...')

        training_set = (
            read_image_file(os.path.join(self.raw_folder, 'train-images-idx3-ubyte')),
            read_label_file(os.path.join(self.raw_folder, 'train-labels-idx1-ubyte'))
        )
        test_set = (
            read_image_file(os.path.join(self.raw_folder, 't10k-images-idx3-ubyte')),
            read_label_file(os.path.join(self.raw_folder, 't10k-labels-idx1-ubyte'))
        )
        with open(os.path.join(self.processed_folder, self.training_file), 'wb') as f:
            torch.save(training_set, f)
        with open(os.path.join(self.processed_folder, self.test_file), 'wb') as f:
            torch.save(test_set, f)

        print('Done!')

    def extra_repr(self) -> str:
        return "Split: {}".format("Train" if self.train is True else "Test")


class FashionMNIST(MNIST):
    """`Fashion-MNIST <https://github.com/zalandoresearch/fashion-mnist>`_ Dataset.

    Args:
        root (string): Root directory of dataset where ``FashionMNIST/processed/training.pt``
            and  ``FashionMNIST/processed/test.pt`` exist.
        train (bool, optional): If True, creates dataset from ``training.pt``,
            otherwise from ``test.pt``.
        download (bool, optional): If true, downloads the dataset from the internet and
            puts it in root directory. If dataset is already downloaded, it is not
            downloaded again.
        transform (callable, optional): A function/transform that  takes in an PIL image
            and returns a transformed version. E.g, ``transforms.RandomCrop``
        target_transform (callable, optional): A function/transform that takes in the
            target and transforms it.
    """
    mirrors = [
        "http://fashion-mnist.s3-website.eu-central-1.amazonaws.com/"
    ]

    resources = [
        ("train-images-idx3-ubyte.gz", "8d4fb7e6c68d591d4c3dfef9ec88bf0d"),
        ("train-labels-idx1-ubyte.gz", "25c81989df183df01b3e8a0aad5dffbe"),
        ("t10k-images-idx3-ubyte.gz", "bef4ecab320f06d8554ea6380940ec79"),
        ("t10k-labels-idx1-ubyte.gz", "bb300cfdad3c16e7a12a480ee83cd310")
    ]
    classes = ['T-shirt/top', 'Trouser', 'Pullover', 'Dress', 'Coat', 'Sandal',
               'Shirt', 'Sneaker', 'Bag', 'Ankle boot']


class KMNIST(MNIST):
    """`Kuzushiji-MNIST <https://github.com/rois-codh/kmnist>`_ Dataset.

    Args:
        root (string): Root directory of dataset where ``KMNIST/processed/training.pt``
            and  ``KMNIST/processed/test.pt`` exist.
        train (bool, optional): If True, creates dataset from ``training.pt``,
            otherwise from ``test.pt``.
        download (bool, optional): If true, downloads the dataset from the internet and
            puts it in root directory. If dataset is already downloaded, it is not
            downloaded again.
        transform (callable, optional): A function/transform that  takes in an PIL image
            and returns a transformed version. E.g, ``transforms.RandomCrop``
        target_transform (callable, optional): A function/transform that takes in the
            target and transforms it.
    """
    mirrors = [
        "http://codh.rois.ac.jp/kmnist/dataset/kmnist/"
    ]

    resources = [
        ("train-images-idx3-ubyte.gz", "bdb82020997e1d708af4cf47b453dcf7"),
        ("train-labels-idx1-ubyte.gz", "e144d726b3acfaa3e44228e80efcd344"),
        ("t10k-images-idx3-ubyte.gz", "5c965bf0a639b31b8f53240b1b52f4d7"),
        ("t10k-labels-idx1-ubyte.gz", "7320c461ea6c1c855c0b718fb2a4b134")
    ]
    classes = ['o', 'ki', 'su', 'tsu', 'na', 'ha', 'ma', 'ya', 're', 'wo']


class EMNIST(MNIST):
    """`EMNIST <https://www.westernsydney.edu.au/bens/home/reproducible_research/emnist>`_ Dataset.

    Args:
        root (string): Root directory of dataset where ``EMNIST/processed/training.pt``
            and  ``EMNIST/processed/test.pt`` exist.
        split (string): The dataset has 6 different splits: ``byclass``, ``bymerge``,
            ``balanced``, ``letters``, ``digits`` and ``mnist``. This argument specifies
            which one to use.
        train (bool, optional): If True, creates dataset from ``training.pt``,
            otherwise from ``test.pt``.
        download (bool, optional): If true, downloads the dataset from the internet and
            puts it in root directory. If dataset is already downloaded, it is not
            downloaded again.
        transform (callable, optional): A function/transform that  takes in an PIL image
            and returns a transformed version. E.g, ``transforms.RandomCrop``
        target_transform (callable, optional): A function/transform that takes in the
            target and transforms it.
    """
    # Updated URL from https://www.nist.gov/node/1298471/emnist-dataset since the
    # _official_ download link
    # https://cloudstor.aarnet.edu.au/plus/s/ZNmuFiuQTqZlu9W/download
    # is (currently) unavailable
    url = 'http://www.itl.nist.gov/iaui/vip/cs_links/EMNIST/gzip.zip'
    md5 = "58c8d27c78d21e728a6bc7b3cc06412e"
    splits = ('byclass', 'bymerge', 'balanced', 'letters', 'digits', 'mnist')
    # Merged Classes assumes Same structure for both uppercase and lowercase version
    _merged_classes = {'c', 'i', 'j', 'k', 'l', 'm', 'o', 'p', 's', 'u', 'v', 'w', 'x', 'y', 'z'}
    _all_classes = set(string.digits + string.ascii_letters)
    classes_split_dict = {
        'byclass': sorted(list(_all_classes)),
        'bymerge': sorted(list(_all_classes - _merged_classes)),
        'balanced': sorted(list(_all_classes - _merged_classes)),
        'letters': ['N/A'] + list(string.ascii_lowercase),
        'digits': list(string.digits),
        'mnist': list(string.digits),
    }

    def __init__(self, root: str, split: str, **kwargs: Any) -> None:
        self.split = verify_str_arg(split, "split", self.splits)
        self.training_file = self._training_file(split)
        self.test_file = self._test_file(split)
        super(EMNIST, self).__init__(root, **kwargs)
        self.classes = self.classes_split_dict[self.split]

    @staticmethod
    def _training_file(split) -> str:
        return 'training_{}.pt'.format(split)

    @staticmethod
    def _test_file(split) -> str:
        return 'test_{}.pt'.format(split)

    def download(self) -> None:
        """Download the EMNIST data if it doesn't exist in processed_folder already."""
        import shutil

        if self._check_exists():
            return

        os.makedirs(self.raw_folder, exist_ok=True)
        os.makedirs(self.processed_folder, exist_ok=True)

        # download files
        print('Downloading and extracting zip archive')
        download_and_extract_archive(self.url, download_root=self.raw_folder, filename="emnist.zip",
                                     remove_finished=True, md5=self.md5)
        gzip_folder = os.path.join(self.raw_folder, 'gzip')
        for gzip_file in os.listdir(gzip_folder):
            if gzip_file.endswith('.gz'):
                extract_archive(os.path.join(gzip_folder, gzip_file), gzip_folder)

        # process and save as torch files
        for split in self.splits:
            print('Processing ' + split)
            training_set = (
                read_image_file(os.path.join(gzip_folder, 'emnist-{}-train-images-idx3-ubyte'.format(split))),
                read_label_file(os.path.join(gzip_folder, 'emnist-{}-train-labels-idx1-ubyte'.format(split)))
            )
            test_set = (
                read_image_file(os.path.join(gzip_folder, 'emnist-{}-test-images-idx3-ubyte'.format(split))),
                read_label_file(os.path.join(gzip_folder, 'emnist-{}-test-labels-idx1-ubyte'.format(split)))
            )
            with open(os.path.join(self.processed_folder, self._training_file(split)), 'wb') as f:
                torch.save(training_set, f)
            with open(os.path.join(self.processed_folder, self._test_file(split)), 'wb') as f:
                torch.save(test_set, f)
        shutil.rmtree(gzip_folder)

        print('Done!')


class QMNIST(MNIST):
    """`QMNIST <https://github.com/facebookresearch/qmnist>`_ Dataset.

    Args:
        root (string): Root directory of dataset whose ``processed``
            subdir contains torch binary files with the datasets.
        what (string,optional): Can be 'train', 'test', 'test10k',
            'test50k', or 'nist' for respectively the mnist compatible
            training set, the 60k qmnist testing set, the 10k qmnist
            examples that match the mnist testing set, the 50k
            remaining qmnist testing examples, or all the nist
            digits. The default is to select 'train' or 'test'
            according to the compatibility argument 'train'.
        compat (bool,optional): A boolean that says whether the target
            for each example is class number (for compatibility with
            the MNIST dataloader) or a torch vector containing the
            full qmnist information. Default=True.
        download (bool, optional): If true, downloads the dataset from
            the internet and puts it in root directory. If dataset is
            already downloaded, it is not downloaded again.
        transform (callable, optional): A function/transform that
            takes in an PIL image and returns a transformed
            version. E.g, ``transforms.RandomCrop``
        target_transform (callable, optional): A function/transform
            that takes in the target and transforms it.
        train (bool,optional,compatibility): When argument 'what' is
            not specified, this boolean decides whether to load the
            training set ot the testing set.  Default: True.
    """

    subsets = {
        'train': 'train',
        'test': 'test',
        'test10k': 'test',
        'test50k': 'test',
        'nist': 'nist'
    }
    resources: Dict[str, List[Tuple[str, str]]] = {  # type: ignore[assignment]
        'train': [('https://raw.githubusercontent.com/facebookresearch/qmnist/master/qmnist-train-images-idx3-ubyte.gz',
                   'ed72d4157d28c017586c42bc6afe6370'),
                  ('https://raw.githubusercontent.com/facebookresearch/qmnist/master/qmnist-train-labels-idx2-int.gz',
                   '0058f8dd561b90ffdd0f734c6a30e5e4')],
        'test': [('https://raw.githubusercontent.com/facebookresearch/qmnist/master/qmnist-test-images-idx3-ubyte.gz',
                  '1394631089c404de565df7b7aeaf9412'),
                 ('https://raw.githubusercontent.com/facebookresearch/qmnist/master/qmnist-test-labels-idx2-int.gz',
                  '5b5b05890a5e13444e108efe57b788aa')],
        'nist': [('https://raw.githubusercontent.com/facebookresearch/qmnist/master/xnist-images-idx3-ubyte.xz',
                  '7f124b3b8ab81486c9d8c2749c17f834'),
                 ('https://raw.githubusercontent.com/facebookresearch/qmnist/master/xnist-labels-idx2-int.xz',
                  '5ed0e788978e45d4a8bd4b7caec3d79d')]
    }
    classes = ['0 - zero', '1 - one', '2 - two', '3 - three', '4 - four',
               '5 - five', '6 - six', '7 - seven', '8 - eight', '9 - nine']

    def __init__(
            self, root: str, what: Optional[str] = None, compat: bool = True,
            train: bool = True, **kwargs: Any
    ) -> None:
        if what is None:
            what = 'train' if train else 'test'
        self.what = verify_str_arg(what, "what", tuple(self.subsets.keys()))
        self.compat = compat
        self.data_file = what + '.pt'
        self.training_file = self.data_file
        self.test_file = self.data_file
        super(QMNIST, self).__init__(root, train, **kwargs)

    def download(self) -> None:
        """Download the QMNIST data if it doesn't exist in processed_folder already.
           Note that we only download what has been asked for (argument 'what').
        """
        if self._check_exists():
            return
        os.makedirs(self.raw_folder, exist_ok=True)
        os.makedirs(self.processed_folder, exist_ok=True)
        split = self.resources[self.subsets[self.what]]
        files = []

        # download data files if not already there
        for url, md5 in split:
            filename = url.rpartition('/')[2]
            file_path = os.path.join(self.raw_folder, filename)
            if not os.path.isfile(file_path):
                download_url(url, root=self.raw_folder, filename=filename, md5=md5)
            files.append(file_path)

        # process and save as torch files
        print('Processing...')
        data = read_sn3_pascalvincent_tensor(files[0])
        assert(data.dtype == torch.uint8)
        assert(data.ndimension() == 3)
        targets = read_sn3_pascalvincent_tensor(files[1]).long()
        assert(targets.ndimension() == 2)
        if self.what == 'test10k':
            data = data[0:10000, :, :].clone()
            targets = targets[0:10000, :].clone()
        if self.what == 'test50k':
            data = data[10000:, :, :].clone()
            targets = targets[10000:, :].clone()
        with open(os.path.join(self.processed_folder, self.data_file), 'wb') as f:
            torch.save((data, targets), f)

    def __getitem__(self, index: int) -> Tuple[Any, Any]:
        # redefined to handle the compat flag
        img, target = self.data[index], self.targets[index]
        img = Image.fromarray(img.numpy(), mode='L')
        if self.transform is not None:
            img = self.transform(img)
        if self.compat:
            target = int(target[0])
        if self.target_transform is not None:
            target = self.target_transform(target)
        return img, target

    def extra_repr(self) -> str:
        return "Split: {}".format(self.what)


def get_int(b: bytes) -> int:
    return int(codecs.encode(b, 'hex'), 16)


def open_maybe_compressed_file(path: Union[str, IO]) -> Union[IO, gzip.GzipFile]:
    """Return a file object that possibly decompresses 'path' on the fly.
       Decompression occurs when argument `path` is a string and ends with '.gz' or '.xz'.
    """
    if not isinstance(path, torch._six.string_classes):
        return path
    if path.endswith('.gz'):
        return gzip.open(path, 'rb')
    if path.endswith('.xz'):
        return lzma.open(path, 'rb')
    return open(path, 'rb')


SN3_PASCALVINCENT_TYPEMAP = {
    8: (torch.uint8, np.uint8, np.uint8),
    9: (torch.int8, np.int8, np.int8),
    11: (torch.int16, np.dtype('>i2'), 'i2'),
    12: (torch.int32, np.dtype('>i4'), 'i4'),
    13: (torch.float32, np.dtype('>f4'), 'f4'),
    14: (torch.float64, np.dtype('>f8'), 'f8')
}


def read_sn3_pascalvincent_tensor(path: Union[str, IO], strict: bool = True) -> torch.Tensor:
    """Read a SN3 file in "Pascal Vincent" format (Lush file 'libidx/idx-io.lsh').
       Argument may be a filename, compressed filename, or file object.
    """
    # read
    with open_maybe_compressed_file(path) as f:
        data = f.read()
    # parse
    magic = get_int(data[0:4])
    nd = magic % 256
    ty = magic // 256
    assert 1 <= nd <= 3
    assert 8 <= ty <= 14
    m = SN3_PASCALVINCENT_TYPEMAP[ty]
    s = [get_int(data[4 * (i + 1): 4 * (i + 2)]) for i in range(nd)]
    parsed = np.frombuffer(data, dtype=m[1], offset=(4 * (nd + 1)))
    assert parsed.shape[0] == np.prod(s) or not strict
    return torch.from_numpy(parsed.astype(m[2], copy=False)).view(*s)


def read_label_file(path: str) -> torch.Tensor:
    with open(path, 'rb') as f:
        x = read_sn3_pascalvincent_tensor(f, strict=False)
    assert(x.dtype == torch.uint8)
    assert(x.ndimension() == 1)
    return x.long()


def read_image_file(path: str) -> torch.Tensor:
    with open(path, 'rb') as f:
        x = read_sn3_pascalvincent_tensor(f, strict=False)
    assert(x.dtype == torch.uint8)
    assert(x.ndimension() == 3)
    return x